Data sources
Data were retrieved from the regional healthcare utilization (HCU) databases of Lombardy Region, the most populated region in Northern Italy, accounting for about 10 million inhabitants (17% of the national population). All Italian citizens have equal access to health care services as part of the National Health Service (NHS), and in Lombardy this has been associated with an automated system of databases to collect a variety of information on residents who receive NHS assistance (NHS beneficiaries), diagnoses and procedures of inpatients of public or private hospitals (coded according to the ICD-CM-9 codes), outpatient drug prescriptions (coded according to the Anatomical Therapeutic Chemical (ATC) Classification System), specialist visits and diagnostic examinations reimbursable by the NHS. In addition, a database reporting the Certificates of Delivery Assistance (i.e., the so called CeDAP registry) providing detailed information on the mother’s socioeconomic traits, as well as detailed information regarding course of pregnancy, labor (e.g. induction), delivery (e.g. mode and timing), health of newborn (e.g. birth weight, Apgar score, presence of congenital malformation or major defect) and causes of perinatal mortality where applicable. A deterministic record linkage between databases through a unique identification code systematically used for all databases allows the identification of large and unselected birth cohorts and the possibility of establishing relevant traits and care pathways of mothers and newborns. 11 To preserve privacy, each identification code is automatically anonymized and the inverse process is allowed only to the Regional Health Authority on request from judicial authorities. Diagnostic, therapeutic and procedural codes used for the current study are given in Supplementary material (Table S1).
As previously underlines the analysis are based on anonymized data: no permission by Ethic Committees is necessary to collect and analyze these data.
Study cohort and PAC definition
All the deliveries in Lombardy between 1st January 2008 and 30th June 2017 from women who (i) were beneficiaries of NHS and resident in Lombardy (at least since three years before the conception date and one year after the delivery date), (ii) were aged 12 to 55 years at delivery, (iii) had 22 to 42 weeks of gestation, and (iv) did not have an ICD-9 code related to cancer in the three years prior to the conception date were identified from CeDAP database. Among these, we excluded deliveries of newborns with very low birth weight (i.e., less than 400 g), deliveries with missing values on Apgar score, birth weight and vitality status, deliveries which did not match to a hospital ICD-9 code related to childbirth, and those in which the infant could not be linked to the mother because of a missing identification code.
Information on cancer diagnosis was obtained from the inpatient database. Accordingly, women with PAC were defined as those having an ICD-9 code of cancer, either in the main or in secondary diagnoses, during the period from conception to delivery (i.e., during pregnancy) or during the following 12 months (i.e. post-pregnancy). Cancer cases reported as secondary diagnosis, for which the main diagnosis was unrelated to cancer, were excluded.
Each woman with PAC was matched to four women randomly selected from those cancer-free, having the same maternal age and year of delivery. PAC and cancer-free women were also compared for socio-demographic characteristics (including nationality, marital status, education and employment), type of conception (i.e. spontaneous or assisted), multiple births, parity and previous history of type 2 diabetes and hypertension (detected in the three years before the date of conception).
Outcomes of interest
Information on pregnancy, delivery and newborn were retrieved from the CeDAP database. Adverse obstetric outcomes included type of labor (induced or no labor due to elective cesarean section vs spontaneous), mode of delivery (cesarean section vs vaginal or instrumental) and timing of delivery (preterm, i.e. 37 gestation weeks or less, vs at term 12). Adverse neonatal outcomes included small for gestational age (SGA) newborn, low Apgar score at 5 minutes (7 or less 13), perinatal death and major malformations diagnosed before discharge of the newborn. SGA was defined as having a birth weight below the 10th percentile for gestational age, according to the sex-specific Italian reference curve for normal fetal growth. 14
Statistical analyses
Descriptive statistics were used to summarize characteristics of PAC and cancer-free women. Differences on categorical variables between the two groups were tested by using absolute standardized differences. Numerical variables were compared between two groups by using the t-test for independent samples.
The overall prevalence of PAC in our study cohort was calculated by dividing the observed number of PAC by the total number of deliveries. Prevalence were further stratified by cancer site. Log-binomial regression models were fitted to estimate the prevalence ratio (PR) and the corresponding 95% confidence interval (CI) of each perinatal outcome among PAC and cancer-free women. Other than the crude PR, adjusted PR (aPR) were also estimated, with allowance for socio-demographic characteristics (nationality, marital status, education and employment), type of conception (i.e. spontaneous of assisted), multiple births, parity and previous history of type 2 diabetes and hypertension. Given the low number of observed cases for some of the outcomes considered, we could not adjust the model for all the aforementioned covariates. Thus, we first used a multivariate logistic regression model for estimating the probability of having a diagnosis of PAC, given the characteristics reported below, and then we used this probability as adjustment covariate in the log-binomial regression model. 15 Moreover, because of the potential correlation of women contributing to more than one birth during the considered period, the models were fitted using generalized estimating equations (GEE) for correlated observations with a log link. 16 Finally, because socio-demographic characteristics were missing for some women (<3%), multiple imputations were applied by using the fully conditional specification (FCS) method, imputing 20 datasets. 17,18
All the analysis were performed among strata of timing of PAC diagnosis (i.e. during pregnancy and post-pregnancy). The frequencies of perinatal outcomes were also reported separately for the three most common cancer sites (i.e. breast, thyroid cancer and lymphoma). Chi-squared test or Fisher’s exact test was used to evaluate differences between cancer sites.
All analyses were performed using the Statistical Analysis System Software (version 9.4; SAS Institute, Cary, NC, USA). Statistical significance was set at the 0.05 level. All p-values were two-sided.